ancel keys and the seven countries study · the seven countries study (scs) is an observational...
TRANSCRIPT
1
Ancel Keys and the Seven Countries Study: An Evidence-based Response to Revisionist Histories
WHITE PAPER
Commissioned by The True Health Initiative http://www.truehealthinitiative.org/
With emphasis on primary source material, historical records, and review/critique by Seven Countries
Study investigators
August 1, 2017
Pett KD1, Kahn J2, Willett WC3, Katz DL4,5
“The whole of anything is never told” Henry James
Acknowledgements: The authors and the True Health Initiative acknowledge gratefully the detailed reviews, critiques, and commentaries of the following researchers directly involved in the Seven Countries Study, and/or in related, subsequent research: Henry Blackburn, MD; Mayo Professor Emeritus University of Minnesota School of Public Health David R. Jacobs, Jr., PhD; Mayo Professor of Public Health, Division of Epidemiology and Community Health, University of Minnesota School of Public Health Martijn B. Katan, PhD; Emeritus Professor of Nutrition, VU University Amsterdam, Department of Health Sciences, the Netherlands Daan Kromhout, PhD, MPH; Professor of Diet, Lifestyle, and Healthy Aging at the University of Groningen, The Netherlands and Adjunct Professor of the University of Minnesota, Minneapolis, USA Alessandro Menotti, MD, PhD; Vice-President, Association for Cardiac Research, Rome, Italy Paolo Emilio Puddu, MD, PhD; Professor of Cardiology, Sapienza University, Rome, Italy and Professor of Physiology, University of Caen, Normandy, France Jaap C. Seidell, PhD; Distinguished Professor of Nutrition and Health and Director, Institute for Health Sciences, Vrije Universiteit and the VUA, the Netherlands
1KatherinePett,MS,MEd,RDN2JoelKahn,MD,FACC,ClinicalProfessor,WayneStateUniversitySchoolofMedicine3WalterWillett,MD,DrPH,ProfessorofEpidemiologyandNutrition,HarvardT.H.ChanSchoolofPublicHealth4DavidKatz,MD,MPHFounder & President, The True Health Initiative, Immediate Past President, The American College of Lifestyle Medicine, Director, Yale University Prevention Research Center 5Correspondingauthor:[email protected]:[email protected]
2
… Competing Interests This work was a voluntary contribution by the authors to the True Health Initiative, and was uncompensated. The authors have no competing interests to disclose.
3
Table of Contents
ABSTRACT 4
EXECUTIVESUMMARY 5
SEVENCOUNTRIES:ANOVERVIEW 9SEVENCOUNTRIES:DESIGNANDMAINOUTCOMES 11SELECTIONOFCOUNTRIES:DIETANDDIPLOMACY 13SELECTIONOFCOUNTRIES:STATISTICS 14PRIMARYFINDINGSOFTHESTUDY 17FINDINGSOFDIETANDDISEASE 17
BEGINNINGTHECONTROVERSY 20ADDRESSINGCHERRY-PICKINGACCUSATIONS:HISTORYOFTHE6-COUNTRYGRAPH 22THESIX-COUNTRYGRAPH 24YERUSHALMYANDHILLEBOE’SRESPONSE 26INFERRINGCAUSALITY 30THEEFFECTS 30THEEXCLUSIONOFFRANCE 32FRANCEWASNOTEXCLUDED 32THEFRENCHPARADOXINTHE1950S:POORRECORDSANDPOORDIAGNOSES 34THESEVENCOUNTRIESSTUDY:DIETARYDATAINGREECE 37DIETARYDIFFERENCESDURINGLENT 37THEROCKEFELLERFOUNDATIONSTUDY 40SUGARVS.SATURATEDFATINHEARTDISEASE 43MECHANISTICINVOLVEMENTINHEARTDISEASE 44AFOLLOW-UPIN1999 46
STRENGTHSANDLIMITATIONSOFTHESEVENCOUNTRIESSTUDY 48
CONCLUSIONS 51
EDITORIALEPILOGUE: 54
THELEGACYOFANCELKEYSANDTHESCS 54
REFERENCES 58
4
ABSTRACT
Disparagements of the methods, intentions, and conclusions of the Seven Countries
Study are currently much in vogue. They populate books and on-line commentary, and
figure prominently in prevailing dietary trends and debates. Critics frequently point out
alleged flaws in the seminal study in order to contest its primary dietary finding, that
saturated fat was correlated with heart disease, and call into question subsequent nutrition
research. This paper was commissioned by the True Health Initiative to explore the
historical record and address the popular contentions with primary source material and
related work, and in consultation with investigators directly involved. Popular criticisms
directed at the study, and the lead investigator, Ancel Keys, turn out to be untrue when
the primary source material is examined.
5
EXECUTIVE SUMMARY
The Seven Countries Study (SCS) is an observational cohort study, started in 1958, to
examine relationships among lifestyle, biomarkers, and heart disease.1–4 The study was a
massive undertaking requiring cooperation among scientists worldwide. Data collection
and analysis spanned decades and were conducted concurrently with numerous
groundbreaking epidemiological studies, including The Framingham Heart Study.5
Ultimately, SCS suggested a link between dietary intake, specifically saturated fat, and
heart disease. This conclusion, which corroborated other clinical and epidemiological
evidence at the time, generated numerous hypotheses and has since inspired countless
clinical trials.
SCS and its originator, Ancel Keys, have come under fire in recent years, particularly
since Keys’ death in 2004, and often in the context of popular nutrition narratives
claiming that prevailing nutrition science is incorrect. Frequently, these critics believe
strongly that carbohydrate, and not fat, is the source of heart disease and other illness-
and do not allow for the possibility that both macronutrient classes, or specific food
sources of them, might be involved. Though SCS itself never concluded that total fat
intake should be restricted, writers – often bloggers – contend that the work of Ancel
Keys and this study in particular are at fault for low-fat dietary trends, guidelines, and
nutrition policy. As a result, modern critiques of seminal nutrition research often use
SCS, and misperceptions around its methodology, as a focal point.
While continuing reanalysis of all science is part of the self-correcting process of the
scientific method, it is important that these criticisms be based in fact and the
documented historical record. There are, of course, limitations to SCS, which will be
addressed in this paper. However, some narratives in the form of revisionist histories
proposed to discredit SCS and Keys are untrue upon review of primary sources. Many of
these narratives have become widely accepted by means of frequent repetition,
particularly since the advent of social media. These criticisms include four primary
allegations:
6
1. Countries were selected and excluded based on desired outcome.
Keys and his team are often accused of picking only countries they believed
would prove their theories and that France was purposefully excluded for this
reason.
While it is true that selection of cohorts was non-random, selection was based on
practicality and dietary variation.1–4 Allegations suggesting that SCS researchers
chose locations where they already knew the outcomes are clearly false based on
review of primary source material, the relevant timelines, and direct questioning
of investigators.
2. France was purposefully excluded
France, which had particularly poor dietary data, was not excluded from the
study.6,7 A French representative was present during the SCS pilot study in
Nicotera, Italy, but ultimately researchers from France decided not to participate. 2,3,8,9
Critics attest that France was excluded because researchers were aware of the
“French Paradox”, but this concept represents an anachronism, since the
information and associations being used today to allege bias were simply not
available at the time the SCS was being designed and implemented.10
3. Dietary data in Greece taken during Lent introduced a distortion.
Some critics believe that nutrient intake, particularly in Greece, was inaccurate
due to dietary surveys being administered during Lent.11
In fact, dietary data were purposefully collected during Lent in order to account
for important seasonal variation in intake.12 Researchers, who published average
7
intake of each survey period in dietary tables, found that there were no
meaningful differences in macronutrient or total energy intake during Lent versus
other times in the two Greek cohorts. These dietary findings from Greece were
consistent with an unaffiliated dietary study completed in Crete in 1948 and
published in 1953.13,14
4. Sugar was not considered as a possible contributor to coronary heart disease.
Critics contend that SCS did not appropriately address sugar intake in the
analysis. Some also suggest that a later re-analysis of SCS data found sugar was
more strongly associated with heart disease than saturated fat.15
The 1980 SCS monograph, an in-depth analysis of 10-year findings, did examine
the association of sucrose and indeed found it to be associated with heart disease.3
However, the association disappeared when saturated fat was also added into the
statistical model, suggesting that sugar was associated with heart disease mostly
due to its strong correlation with saturated fat in the diet. In contrast, the
association between saturated fat and heart disease persisted after adjusting for
sugar, suggesting it to be the primary and more significant factor. The 1999
reanalysis, contrary to one retelling, did not find that sugar was more highly
associated with coronary heart disease than saturated fat; its findings were
consistent with earlier SCS publications.16
This paper examines each of these assertions using evidence from primary sources,
historical records, and observations from SCS investigators directly involved, and
assesses each contention in historical and scientific context.
First, the paper provides the historical context and description of the SCS: how it was
conducted, concurrent epidemiological studies, and the study findings reported through
1986. Next, the paper examines and assesses each of the most common criticisms one by
one. Conclusions regarding the popular narratives (i.e., revisionist histories) are drawn
8
with direct reference to the relevant evidence. Limitations to the SCS are summarized as
well.
Finally, in an epilogue/editorial comment, the legacy of the SCS and the lifelong
work of Ancel Keys are briefly considered. What role, if any, did Ancel Keys play in the
“low fat” era of nutrition policy? What conclusions were reached by Keys regarding
saturated fat, dietary cholesterol, other dietary factors, serum cholesterol, components of
the lipid panel (e.g., LDL, HDL) and heart disease- and how do these compare with the
prevailing understanding of experts today? What is known about how Keys changed his
opinions about these matters over time, and the sources used to inform such change?
This paper does not espouse or promote any dietary advice; it is intended only to
present a historically accurate account of well-documented work and redress
misrepresentations of that work. Further, but for brief mention in the epilogue, this paper
does not address how the science and research detailed here may have been co-opted and
misinterpreted by various entities in the service of various motivations not directly
accountable to the epidemiologic data. Industry and government actions during this
period represent a complex and contentious topic beyond the scope of this White Paper.
9
SEVEN COUNTRIES: AN OVERVIEW
The Seven Countries Study began in 1958 to measure dietary habits, biomarkers, and
lifestyles of 12,763 men in seven countries: the United States, Italy, Finland, Greece, the
Netherlands, Japan, and two nations of the former country of Yugoslavia (now Croatia
and Serbia). The aim was to assess connections of both lifestyle and anthropometric
measures to the risk of developing or dying from coronary heart disease.1–4,12,17 The
subjects of the research were 16 cohorts of men aged 40-59. Eleven cohorts lived in rural
villages, two cohorts were railroad workers — one based in the United States and another
in Rome, Italy — one cohort was comprised of Belgrade University faculty in Serbia, one
was comprised of workers of an agro-industrial cooperative in Zrenjanin, Serbia, and
another lived in the country town of Zutphen, the Netherlands.1–3
In 1957, preparation for SCS began with a pilot study in Nicotera, Italy, a town that
ultimately was not to be included as a cohort in the study.1,3,9 This is where researchers
developed and tested their standardized methods for recruitment and enrollment and
biomarker measurements to ensure collection would be uniform across all sites.
Researchers representing all sites were present at the pilot study. Nicotera served as a
dress rehearsal for the main study, which began in 1958.3,18
In each town, investigators enrolled eligible men and collected baseline data
including blood cholesterol, blood pressure, BMI, and electrocardiograms (ECG).2,3 They
collected participants’ medical and smoking histories, and they classified level of
physical activity based on profession. Diet records were taken in representative
subsamples of cohorts as described below.
For SCS, researchers developed new procedures to improve the accuracy and
reliability of any medical diagnoses. In previous studies that relied on national disease
and death data comparisons, differences in diagnostic criteria for heart disease accounted
for much of the variability between countries.10,19 To avoid this, the investigators
developed strict codes for categorizing diagnosis.3,20–22 All ECGs were sent to the
10
University of Minnesota for analysis and classification to reduce variability in data
interpretation. The Minnesota Code was developed to reduce observer error and provide
strict guidelines so type and incidence of coronary heart disease were uniformly
categorized.
Researchers were similarly concerned with collecting consistent dietary data. In all
but the US and Italian railroad cohorts, researchers collected thorough seven-day diet
recall diaries;3,4 a representative subsample of 30-50 men was chosen from each cohort to
complete intensive seven-day weighed food records. Under dietitian supervision, all food
and drinks consumed were weighed. In addition, duplicate samples of all foods eaten
were collected, freeze dried, and sent to the University of Minnesota for chemical
analysis.4,12 Both the results from dietary recall data and weighed food sampling were
published and compared in order to validate intake measures.12,23 In the US, intake was
estimated using 24-hour recalls and food frequency questionnaires. For the Italian
railroad cohort, seven-day diet records and food frequency data were collected.4
Achieving a high follow-up rate among participants was a primary concern during
study design, and the choice of measurements was made with this in mind. Oral glucose
tolerance was suggested as part of an initial protocol, but was excluded due to expense
and concern that participants might not return if visits lasted an additional hour.3
Beginning in 1970, glucose tolerance tests were carried out in some cohorts, notably
Zutphen, Netherlands, over the remaining course of the study.18
At five and 10 years, all cohorts were revisited — except the two Japanese cohorts
were not examined at year 5 and the U.S. Railroad cohort, which was not examined at
year 10 due to lack of funding — and biomarkers and electrocardiograms were taken
again.3 New cases of heart disease were diagnosed and death records from all cohorts,
including those for the railroad workers cohort in the U.S., were carefully combed.
Deceased participants’ causes of death were examined and validated by the research
team. Publications following up on the SCS cohorts after the 10-year mark examined
coronary heart disease mortality and all cause mortality.4,16,18
11
As with coronary heart diagnoses, similar care was taken in verifying official cause of
death. Death certificates for each individual, rather than official mortality statistics, were
used and in more than half of deaths, researchers corroborated cause of death by
interviewing a participant’s family, physicians, and/or obtaining hospital records. Even
when only a death certificate was available, a critical review was made to define cause of
death according to standardized criteria.18
Three book-length study analyses were published in 1966, 1970 and 1980.1–3 The 10-
year (1980) and 15-year (1986) follow-up publications are important sources for the
present paper; many other details can be found in the voluminous literature generated by
the SCS.3,4 The great number, and length, of SCS publications can prevent readers from
grasping the scope and details of the study, since each paper inevitably presents only a
small part of the whole story. Keys, and the large team of authors representing all of the
participating nations, tracked correlations, evaluated outcomes, and determined what
conclusions could be drawn from decades of evidence. SCS primarily examined the
relationship among cohorts between diet and coronary heart disease, and secondarily,
correlations within cohorts among lifestyle, biomarkers, and coronary heart disease.2–4,18
Seven Countries: Design and Main Outcomes
The Seven Countries Study is a type of observational study known as a “prospective
cohort study.” For a cohort study to examine outcomes such as coronary heart disease, it
must enroll participants who do not possess the condition at baseline and then collect data
on these participants and cohorts over time. As time goes by, researchers assess the
incidence of and mortality from a given disease within and/or across the study
populations. Researchers then track whether baseline measurements, like high blood
pressure, are associated with differing propensities for disease development and
mortality. Because researchers only observe participants and do not introduce dietary or
lifestyle interventions, this is known as an “observational” study or specifically “cohort
12
study.” A cohort study that follows participants over an extended period of time can also
be referred to as “longitudinal” or “prospective.”
Perhaps the most famous study of this type is The Framingham Heart Study, which
followed select citizens of Framingham, MA, and was one of the first studies to show the
damages caused by cigarette smoking.5,24,25 To provide some historical context, The
Framingham Heart Study was merely 8 years old at the start of the Seven Countries
Study; the first major findings were published in 1957, a year before official data-
collection in SCS would begin.24
Numerous cohort studies examining relationships between lifestyle and coronary
heart disease were in their early years at the time of SCS. The Minnesota Business and
Professional Men Study, also by Keys and his Minnesota colleagues, had started in
1947.26 In a table of cohort studies predating SCS, Keys noted that on the West Coast, a
cohort study in Los Angeles civil servants started in 1950 and one in San Francisco
longshoremen started in 1951.27 In Albany, NY, similar observational studies began on
civil servants in 1953.3 Other major observational studies would continue to form
throughout the ‘50s and ‘60s and sample sizes would trend upward.3
Unlike many prospective cohort studies, which are performed in a limited
geographical area, the cohorts in SCS were ecologic units. Many of the aspects of the
SCS, including diet, were analyzed as an “ecological” study, meaning that Keys et al.
examined correlations between the mean values of data points of the 16 geographic
cohorts, as well as comparing the data from each individual.3 An ecological study looks
at the associations between exposure variables, and outcome variables, at the level of the
sub-population. A semi-ecologic study also collects and analyzes data among individuals;
thus, depending on the focus of study, SCS is an “ecologic” or a “semi-ecologic” study
(see “Strengths and Limitations,” page 48).
13
Selection of Countries: Diet and Diplomacy
Running a multinational cohort study in 1956 presented a series of challenges. Only a
little more than a decade past World War II, there were diplomatic and practical concerns
for country selection. In order to participate, funding needed to be secured for each
cohort, a precluding barrier for some countries. Indeed, SCS was unable to sustain
funding for its American cohort and so disease incidence of the US cohort, assessed by
ECG and clinical examination, was dropped as an outcome after the five-year follow-
up.1,3
Competency and familiarity with each region were other considerations. The
University of Minnesota team attempted to work with researchers and in regions with
which their international colleagues had knowledge and contacts, and where they were
welcome.3 Cooperation from village leaders and citizens was essential for adequate
follow-up, which required the presence of scientists native to each of the included
countries. Keys and his team benefitted from existing international contacts that assisted
in scouting and suggesting locations. Many of the small communities chosen had
participation rates of greater than 90%, an impressive feat requiring substantial local buy-
in. The lowest participation rate for coronary heart disease incidence, still comparatively
high among observational studies, was among the American railroad workers who were
concerned that their jobs would be at stake if researchers discovered poor health.3
Follow-up for total mortality was near 100% for the first 15 years.18 Follow-up for vital
status and cause of death was nearly complete for most cohorts for 25 years, and for
select cohorts, for 50 years.
14
Selection of Countries: Statistics
Researchers carefully selected populations across a wide variation of dietary patterns,
but with low variance of other characteristics to limit confounding. The rural cohorts in
SCS were identical in age distribution and gender, and highly similar in level of
education, socioeconomic status, and laborer lifestyle, but not in diet.1–4 This ideally
reduced confounding when comparing cohort means, while maximizing variability in
cohort diets and coronary heart disease incidence.
At the time of SCS, all epidemiological studies examining health outcomes in relation
to diet in single populations encountered a nearly insurmountable practical and statistical
problem. The day-to-day variation in the diet of a single individual (called intra-
individual variation) was of nearly the same magnitude as variation in diet between
individuals (inter-individual) of the same culture.3,28–30 In order to be reasonably
confident of a difference between one person’s diet and his neighbors’, dietary surveys
would need to have been repeated upwards of seven times.3 This approach was
impractical, prohibitively expensive and, even then, unlikely to discover major
differences among individuals eating a relatively homogenous diet without a much larger
sample size.
For these reasons, Keys thought it was imperative to compare disease rates where the
average diets of populations varied over a wide range in order to see possible
relationships between nutrient intake and chronic disease risk.1–3,18 Consider that in a
population where everyone smokes, it would be nearly impossible to find smoking
associated with increased risk of lung and mouth cancers, even though true. A
measurable difference in outcomes requires a discernible difference in exposure.
Hence, the choice of populations with divergent dietary patterns was crucial to the
early study of possible causes of heart attack. Historically, much of the initial information
that suggests diet’s impact on health has come from observations of populations with
outlying diets and disease incidence, including the classic observations in China and Java,
15
Seventh Day Adventists in Loma Linda, and smaller but more recent observations among
the Maasai in Kenya and northern Tanzania, and most recently among the Tsimané of
Bolivia.31–34
Of the seven countries, the United States and the Netherlands each provided a single
cohort, the former Yugoslavia (now Croatia and Serbia) provided five, Italy three, and
Japan, Greece, and Finland each contributed two.1–4 In most cases, cohorts within each
country were selected because of purported differences in eating habits, thus potentially
providing dietary variation between a region’s cohorts. In the analysis, comparisons were
therefore made between individual cohorts as well as between countries.
The United States was an obvious choice because the research team was based in
Minnesota.3 Researchers and public health officials were interested in discovering factors
linked to the relatively high incidence of coronary heart disease plaguing middle-aged
men in the U.S.
Japan had an extremely low rate of reported coronary heart disease and a national diet
uniquely low in fat, though it was unclear whether particular health outcomes reported
were due to diet, the Japanese tendency to record sudden deaths as cerebrovascular, or
both.1 The Japanese cohort in Tanushimaru was a farming community, while the
Ushibuka cohort was a fishing community with a diet higher in seafood.18
North Karelia, Finland, was known for high-fat diets and allegedly had extremely
high rates of coronary heart disease. The West Finland cohort included men in two rural
villages, where coronary heart disease was purportedly lower than in East Finland.18
The Greece cohorts in Crete and Corfu were included because they reported low
mortality from coronary heart disease, but high intakes of fat, particularly olive oil.3,18
Sites in Croatia (former Yugoslavia) were chosen with help from a native colleague,
Ratko Buzina, who provided staff for data collection in these regions.3 This allowed
16
researchers to minimize costs and achieve high survey response rates. Interestingly, the
five sites in former Yugoslavia were characterized by supposedly different dietary
patterns. The Dalmatia cohort was known for a high vegetable Mediterranean-style diet,
the Slavonia cohort had a more traditional central European diet high in animal
products.18,35 The three Serbian cohorts were chosen based on varying socioeconomic
status and reportedly had similar diets; Zrenjanin included men working on an
agricultural cooperative; Belgrade, the faculty at the University of Belgrade; and Velika
Krsna, farmers in the rural village south of Belgrade.18,35
The rural regions of Italy were chosen for similar reasons: varied diets between
cohorts. The Rome railroad workers provided the “Italian counterpart” to the U.S.
railroad cohort. Crevalcore, in northern Italy, purportedly had a diet higher in animal fat
while Montegiorgio had a diet more traditionally “Mediterranean.”18 The Netherlands
was included because the Dutch government nutritionists were collegial and curious to
join the pioneering comparisons among cultures.3,36
Notably, Keys and his team wanted to include areas with differing rates of coronary
heart disease, but relied on uncertain estimates by local experts of the prevalence of
disease, since government reported death rates at the time were unreliable indicators.7
Much of the modern criticism of the SCS is aimed at the selection or non-selection of
various regions, a topic that will be covered in a later section (page 22). It’s important to
note that country selections aimed to include a wide range of diets so that a statistical
effect could be measured if it existed. Selection was also shaped by the practical
considerations of cost and connections imposed on carrying out research in any capacity.
Additionally, other than preferential addition of developed nations with very diverse diets
to improve the power of the study, lack of interest, preparation, or funding were the only
reasons for excluding countries. Countries with willing researchers and ready funds were
welcome.8
17
This paper focuses primarily on the findings reported in the 1980 and 1986
publications, which evaluated 10- and 15-year coronary heart disease incidence and
mortality, as well as all-cause mortality.3,4
Primary Findings of the Study
Deaths from coronary heart disease were consistent with the data coming from
scattered sources before the study commenced.3 Northern Europe and America
experienced a much higher incidence of and mortality rate from coronary heart disease
than did men of the same age in Southern and Central Europe. Japan was characterized
by very low reported coronary heart disease death rates.3
The finding most debated today was that the average cohort intake of saturated fat (as
percent of calories or energy) was statistically significantly associated with cohort rates
for coronary heart disease3. Total dietary fat was not associated with coronary heart
disease. For example, the Greek cohorts ate about 40% of their daily calories from fat and
the Japanese ate only 10% while both cohorts had very low coronary disease rates.3,4 SCS
was the first systematic study to illustrate this important dichotomy, that low rates of
coronary heart disease and, in fact, of total mortality can be found with low and high total
fat intakes, depending on the nature/sources of the fat and the rest of the eating pattern.3
Findings of Diet and Disease
Keys et al. assessed associations among cohorts using a multivariable linear
regression.3 This method was to evaluate potential ecological correlations between
lifestyle and disease while controlling for potential confounding factors. Simple
correlations can be found by comparing two measurements, but don’t allow researchers
to conclude that the correlation is specific to the dependent and independent variables
examined. This is because there are other circumstances possibly associated with both
major variables that are not accounted for in a simple correlation; these are known as
18
confounders. Adjusting for confounders does not allow researchers to overcome the
primary limitation of observational studies and determine causal relationships, but can
increase confidence that a correlation is valid and meaningful.37
A multivariable regression accounts for confounding factors by controlling for them
within the statistical analysis. With this method, researchers can assess the effect of the
exposure (such as a diet) on the outcome (such as coronary heart disease) including
possible confounders in the equation. By including multiple factors possibly associated
with coronary heart disease including weight, blood pressure, physical activity, and diet,
researchers could more reliably distinguish between true correlation and coincidence. Of
note, however, is that such adjustment is never perfect, and particularly in an ecological
analysis does not serve to establish cause-and-effect.
Keys et al. measured the ecologic association between saturated fat and coronary
heart disease controlling for total caloric intake3. Sugar was included in the model, which
assessed the effect of saturated fat intake when adjusted for sugar intake. Saturated fat
remained statistically significantly associated with heart disease in that model. The
opposite circumstance did not hold true. When sugar was the independent variable,
adjusting for saturated fat intake eliminated any observable association between sugar
consumption and the incidence or mortality from coronary heart disease.3,38 In other
words, the analysis suggested that any variation in heart disease concurrent with sugar
intake was “explained away” by variation in saturated fat intake, whereas variation in
heart disease concurrent with variable saturated fat intake was not explained away by
variation in sugar intake. Saturated fat emerged directly from the data analysis as the
predictor variable of singular apparent importance.3,4
There are numerous claims, mainly on blogs, attesting that SCS did not include sugar
in its statistical models; this is not the case.3,38 However, high intake of refined
carbohydrates is now considered to be a possible contributor to heart disease
development;39,40 this was not evaluated. In SCS, saturated fat intake was correlated with
sugar intake and cohorts with diets higher in saturated fat also consumed more sugar.3
19
Tellingly, saturated fat intake was non-significant when serum (not dietary)
cholesterol was added into the statistical model as a control variable.3 Earlier
experimental studies had clearly demonstrated that saturated fat intake caused increased
serum blood cholesterol when substituted for carbohydrate or compared to
supplementation with mono or polyunsaturated fatty acids.41,42 The significant
relationship between saturated fat intake and heart disease was eliminated when adjusted
for serum cholesterol, which suggested that serum (not dietary) cholesterol was a key
mediator.3 To summarize: Keys et al.’s findings suggested that the correlation between
saturated fat and increased coronary heart disease was due to the correlation between
saturated fat intake and higher serum cholesterol levels, which subsequently affect
coronary heart disease risk. Sugar was analyzed in exactly the same way as saturated fat,
but did not demonstrate a significant association with relevant outcomes independently.3
Findings from the 1986 evaluation of coronary heart disease and all-cause mortality
were consistent with those of the 1980 monograph.3,4 Further, Keys et al. found that
saturated fat intake, specifically the ratio of saturated to monounsaturated fatty acids, was
significantly associated with death from coronary heart disease as well as all-cause
mortality.3,4
Though high serum cholesterol was significantly associated with coronary heart
disease outcomes, at the time of the analysis in 1980, Keys et al. concluded that there was
likely a threshold under which average population cholesterol level might not be
associated with increased risk.3 Later evidence with huge samples found individual
coronary heart disease risk continuous with serum cholesterol, particularly LDL, from
low to high.43
20
BEGINNING THE CONTROVERSY
The SCS was one of many longitudinal observational studies in the mid 20th century
that attempted to discover why coronary heart disease mortality in the United States and
Northern Europe seemed high. Unlike most of those concurrent studies, Keys et al. used
an ecological approach to compare the lifestyles and risk factors among entire
populations and diverse regions across a wide variety of traditional eating patterns.
Why has the study become controversial? Over time, numerous arguments emerged
asserting that Keys’ seminal study was wrong, or worse, that data were altered or selected
to reflect specific outcomes. It is possible that modern critiques of a study that began over
60 years ago led to careless handling of historical facts.
The SCS, like all scientific endeavors, was not without limitations, which will be
summarized below. But attempts to paint the study as “corrupt” or “cherry-picked” lack
merit.
To demonstrate this, we critically examine claims frequently made including:
1. That Keys, et al. selected countries with a specific outcome in mind; that SCS
had data from 22 countries but highlighted only seven.
2. France was deliberately excluded because the SCS research team was aware of
the “French Paradox.”
3. That dietary surveys made in Greece were invalid because they occurred
during Lent.
3. That an analysis of SCS data published in 1999 proved that sugar, and not
saturated fat, was associated with coronary heart disease.
Authors referred to primary literature as well as communication with original
members of the SCS research team to evaluate claims that have become popular on blogs
and in diet books. The authors note that this effort illustrates “Brandolini’s Law:”
21
“The amount of energy needed to refute [misinformation] is an order of magnitude bigger than to produce it.”44
22
Addressing Cherry-picking Accusations: History of the 6-Country Graph
Many critiques start with a cherry-picking argument: Keys had data from 22 countries
and chose only the seven that fit his hypothesis.45 This accusation is startlingly pervasive,
yet demonstrably false.
Where does this argument come from? Misrepresentations of the source of the data
are years old; Keys himself responded to critics who incorrectly thought SCS data came
from national statistics.46 Speakers, bloggers, journalists, and other detractors of SCS
frequently cite this graph:47
Figure 1: From "Atherosclerosis: A problem in newer public health," Journal of the Mount Sinai Hospital, 1953. This image has been reproduced from widely available reproductions accessible online. Permission to reproduce the original has been sought from the publisher, and a response is pending at this time.
And then this one:6
23
Figure 2: From "Fat in the diet and mortality from heart disease: A methodologic note," 1957, New York State Journal of Medicine. Permission to reprint granted by the publisher.
The first graph is from a paper by Ancel Keys, showing a correlation between fat in
the diet, calculated from national disappearance data, and World Health Organization
(WHO) data on heart disease in six countries.47 The second, from a paper by statisticians
Jacob Yerushalmy and Herman Hilleboe, shows the same relationship but with 22
countries.6 Many critics suggest that the second graph eliminates the correlation between
fat intake and heart disease. Though it is easy to see that the positive correlation between
fat intake and heart disease remains in this version, it does appear weaker.
24
The seemingly stark contrast of these graphs side-by-side has become an enduring
argument for the inaccuracy of, and alleged “cherry-picking” of countries for — the SCS.
But neither of these graphs is from the SCS.
The first graph of six nations is from an earlier paper by Ancel Keys in 1953, more
than a decade before the first publications from SCS.47 The second is from a paper
published in 1957, nearly a year before baseline data collection began in SCS cohorts.6
Clearly, neither of these graphs uses the same data as the SCS.
The first graph is often mistakenly attributed to SCS, and critics of the study are not
fastidious about correcting the mistake. The incorrect assertion that SCS started with 22
nations and pared down to the seven that fit Keys’ hypothesis even made it to the
Wikipedia entry for SCS for a time.45 A cursory examination of the primary source
material reveals it to be a false proposition.
Legitimate confusion on the topic is likely due to the passage of time and failure to
review the full historical record. The historical context behind the development of each
graph readily clears up the controversy surrounding this mistaken assertion.
The six-country graph
In 1953, Ancel Keys’ presentation at a symposium, “Recent Advances in Therapy,” at
Mt. Sinai Hospital included the first graph displaying six countries.47 This presentation
was then published as a paper the same year, titled, “Atherosclerosis, a problem in newer
public health.”47 In it, Keys presented a wide-ranging review of the evidence available at
that time, suggesting that atherosclerosis was not simply a natural consequence of aging
and that diet might be related to heart disease outcomes. What was revolutionary about
this was the idea that heart disease could be preventable. This was an intuition Keys
could not prove at the time, though his future work and that of many others, have proven
him right.39,40,47,48
25
In the paper, Keys pointed out that diagnosis of cardiovascular disease, which he
referred to as “degenerative diseases of the heart,” varied greatly but seemed to share
common biological pathways.47 His second point was to show that serum cholesterol
tended to be higher in people with heart disease than in people without it. And thirdly, he
reviewed previous experimental evidence in humans showing that altering the amounts of
fat and dietary cholesterol in the diet affected serum cholesterol levels.47
This, along with the six-country graph, was meant to posit the following thought
process. The mortality for middle-aged men in the United States was higher than in
similarly developed countries; this mortality difference seemed to be due to higher
mortality from coronary heart disease. A higher fat intake was associated with a greater
incidence of heart disease in some countries. His conclusion was to suggest that fat intake
was correlated with coronary heart disease through serum cholesterol level, but that type
of fat might also matter.47
Keys said at the time: “Whether or not cholesterol, etc., are involved, it must be
concluded that dietary fat is somehow associated with cardiac disease mortality, at least
in middle age.” It is interesting to note that at this time, in 1953, Keys primarily posits
that change in risk may be attributable to total fat, rather than saturated fat, in the diet.47
He would later refine his thinking on this topic with data from his own work in human
cross-over dietary experiments, clinical trials, and in SCS, where only saturated fat
proved to be associated with an increase in risk.3,4,41,42 This, according to Keys’ many co-
investigators, was his consistent pattern; he would methodically test the hypotheses he
formed on the basis of provisional data, and evolve his thinking in accord with the data
and findings.49
The six-country graph appeared in a research paper that ran more than 20 pages and
included tables and results from numerous comparative studies showing the differences
in mortality, coronary heart disease mortality, and food disappearance data among
multiple countries, results of experimental data on diet and serum cholesterol, and
observational comparisons of average serum cholesterol in populations with and without
26
heart disease.47 None of this was decisive proof of cause-and-effect, but was the start of a
line of research still yielding new insights, and the basis for Keys’ hypotheses at the time.
Why did Keys choose only six countries for the particular graph? According to Keys,
these six countries represented those that had the most reliable dietary and vital statistics
at the time, in 1952.47 A consistent principle in the Laboratory of Physiological Hygiene
(Keys’ Department at the University of Minnesota) at the time and for many years later
was: the data you talk about have to be of the highest caliber, valid, reliable, and most
relevant to the scientific question.50 Additionally, numerous scientists had noticed that
WWII had changed not just total mortality statistics, but coronary heart disease death
rates as well, and that countries occupied by Germany during the war experienced
significant upheavals to traditional diets that altered their risk factors for heart disease.
“In Norway,” Keys pointed out, “the public health and vital statistics records were well
maintained and it is clear that not long after the national dietary change began, there was
a marked decline in mortality from circulatory disease, particularly arteriosclerotic heart
disease.”47
Though the paper was published in 1953, the data used by Keys in the graph (and
likely the most recent data available) were from 1949, only four years after the end of the
war. Dietary changes in Germany and the countries it occupied lasted longer than the end
of the war and could possibly have introduced bias into a simple correlation.51 In a later
paper, Keys noted “the prolonged influence of the war and its aftermath on the diets of
many countries cannot be ignored.”7,52,53
Yerushalmy and Hilleboe’s response
In 1955, Keys presented to atherosclerosis experts in Geneva the same information, in
a talk that was published as “The relationship of the diet to the development of
atherosclerosis in man.”6,54 Two public health professionals, Jacob Yerushalmy and
Herman Hilleboe, were in attendance. They were not impressed by what they interpreted
as Keys’ confidence that dietary fat was causally linked to heart disease; they felt there
27
was not substantial evidence to make a causal claim.6 Yerushalmy and Hilleboe chided
Keys for altering his “cautiously observed” correlation in 1953 to the more strident
assertion he made in 1955:
“The analysis of international vital statistics shows a striking feature when the national food consumption statistics are studied in parallel. Then it appears that for men aged forty to sixty or seventy, that is, at the ages when the fatal results of atherosclerosis are most prominent, there is a remarkable relationship between the death rate from degenerative heart disease and the proportion of fat calories in the natural diet. A regular progression exists from Japan through Italy, Sweden, England and Wales, Canada and Australia to the United States. No other variable in the mode of life besides the fat calories in the diet is known which shows anything like such a consistent relationship to the mortality rate from coronary or degenerative heart disease.”6
Yerushalmy and Hilleboe, concerned that Keys was overstating a simple correlation,
decided to investigate these claims.
Four years after Keys’ initial paper was published, Yerushalmy and Hilleboe
published a critique in July of 1957, featuring a 22-country graph showing Keys’
correlation substantially weakened.6 In September of 1957, Hilleboe published another
19-page paper focused on the same topic covering similar points.55 Of the 15 tables and
graphs included in Keys’ 1953 paper, which relied on a thorough history of clinical,
experimental, and observational evidence, Yerushalmy and Hilleboe’s July 1957 paper
chose to focus on a single piece of evidence, the six-country graph.6
They took issue with Keys’ selection of countries, since they noted that data on
dietary intake and heart disease mortality existed for 22 countries.6 Yerushalmy and
Hilleboe, however, failed to consider that food intake and heart disease incidence were
probably influenced by the war and Nazi occupation. In fact, numerous countries
included in their graph — Germany, Norway, the Netherlands, France, Denmark, and
Greece — experienced atypical food availability and intake in the years prior to and
directly after the war, the effects of which had been mentioned in Keys’ 1953 paper.7,47,51
28
Yerushalmy and Hilleboe’s criticisms of Keys’ 6-country correlation graphic were to
some extent valid, though the graph represented only a fraction of Keys’ larger
argument.6 First, they pointed out that any simple correlation by itself could not prove a
causal link between an exposure and an outcome, thus introducing the criteria for causal
inference from correlations. In fact, Keys had not claimed causation.47
Next, they questioned the quality and accuracy of heart disease diagnoses among
countries. Some countries were more precise with their diagnoses of heart disease, while
other countries were more likely to file deaths under a broader category of heart-related
deaths.6 These imprecisely categorized deaths were not included in Keys’ initial 1953
six-country assessment, and Yerushalmy and Hilleboe criticized the omission, though it’s
possible Keys considered quality of diagnosis in his selection of the six countries with
“fully comparable dietary and vital statistics.”47 Further, Yerushalmy and Hilleboe had
found another dietary factor more significantly associated with heart disease than fat:
animal protein.6
They took care to point out that their observation shouldn’t suggest animal protein
was truly correlated with increased heart disease since an “apparent association often
proves to be the result of non-pertinent extraneous factors.”6 They intended only to point
out that multiple dietary correlations could be made and that scientists can’t rely on single
associations to assume cause. In fairness, neither of Keys’ previous papers had relied on a
single correlation; they included the results of multiple observational and experimental
studies.47 Nor was causality asserted; Keys simply hypothesized in accord with the
findings, and these hypotheses, in part, were the basis for the SCS.
While Yerushalmy and Hilleboe accurately pointed out the liability of inferring
causation from correlation, their paper was not without major flaws of its own. After the
papers published in July and September of 1957, Keys responded with an editorial in the
Journal of Chronic Disease that November.7
29
According to Keys, Yerushalmy and Hilleboe had used dietary and health data
irrespective of quality; this is something Keys felt could affect results. Secondly, Keys
pointed out that in some of the included countries, dietary data could not possibly reflect
coronary heart disease data because some countries’ populations included had a large
percentage of recent immigrants whose diets were affected by their culture and country of
origin. The dietary data, then, were not reflective of the population being measured and,
thus, could not accurately be correlated with disease incidence.7
Most importantly, Keys pointed out that Yerushalmy and Hilleboe suffered from a
chronological problem – something that would later be codified in the Bradford Hill
criteria.7,56 In his September 1957 paper, Hilleboe used dietary data primarily from 1954
to analyze the relationship with earlier coronary mortality data from 1950-52.55
Keys points out that the foundations of coronary heart disease start years before
disease mortality. It is more likely that a country’s average diet in the mid- to late 1940s
would show an accurate correlation with heart disease deaths in the 1950s than would
concurrent dietary data, and implausible that a country’s average diet after recorded
disease mortality could be a true correlation, particularly if dietary patterns had changed
over time.7
It is also important to note that dietary data of the mid-1950s were not reflective of
dietary data a decade previously in many of these European countries that were gravely
affected by World War II.7 Keys points out that this is particularly the case with the
Netherlands:
“In the data as tabulated by Dr. Hilleboe and Dr. Yerushalmy, the most striking
departure from a good correlation between dietary fat in 1954 and mortality in 1950-
1952 is the case of the Netherlands… There are no reliable records for the war years, but
it is known that food fats available in the Netherlands were greatly reduced from 1940-
1944, that the diet was at the near starvation level… In no case is it proper to suggest
that the deaths in 1950-1952 should reflect the diet of the surviving population in later
30
years.”7 This comment by Keys was in direct response to a related paper published singly
by Hilleboe.55
Inferring causality
Those well-versed in epidemiology will recognize the applicability of Bradford Hill’s
criteria governing the inference of causation from correlation, but it is important to point
out that these criteria, taken for granted today, were not published until the mid 1960s.56
That is over a decade after this debate on saturated fat, and indeed after the Surgeon
General’s Report on Smoking (1964) that largely relied on epidemiological evidence to
infer causation between smoking and lung cancer.57 The academic debates on how
epidemiology should be used to influence public health policy, so eloquently summarized
in the Bradford Hill criteria, had their genesis in debates that took place in the 1950s,
including this one between Keys, and Yerushalmy and Hilleboe.58
The effects
Yerushalmy and Hilleboe, intentionally or otherwise, distorted Keys’ original point
by citing a single piece of data and critiquing it as if it were the basis for his entire line of
reasoning. However, in doing so, they made several good — if pedantic — points about
the limitations of observational evidence for inferring causal relationships.6,55
By the time Yerushalmy and Hilleboe’s paper was published, SCS was already
underway incorporating procedures designed to avoid the exact traps Yerushalmy and
Hilleboe highlighted.3 SCS would standardize the way diet and heart disease were
measured in cohorts. Rigid protocols eliminated much of the variability in local
infrastructure and diagnosis. In-depth surveys and chemical analysis of food samples
increased the exactitude of dietary data compared to relying on a country’s food
availability data. SCS would also tailor methodology to control for what Yerushalmy and
Hilleboe called the “non-pertinent extraneous factors” that create false correlations.3,6
The SCS team selected locations that allowed entire towns to be sampled and enrolled
31
mostly men of the same age and socioeconomic status, which ensured that sampled
populations would be largely similar except with regard to diet and lifestyle, the
independent variables of interest. The study’s prospective design allowed it to enroll
subjects before disease began and carefully tracked the representative diet of each cohort;
giving it a great deal more validity than using national or FAO statistics.3
In conclusion, the famous six-country graph had nothing to do with the Seven
Countries Study and was produced from an earlier analysis that used unrelated and less
robust data. Further, the epidemiological debate between Ancel Keys and Yerushalmy
and Hilleboe concerned issues about the specificity of correlations, which were controlled
for in the later and more precise and extensive data obtained in the SCS. The
presentation of graphs of 6 and 22 countries, juxtaposed to claim cherry-picking in the
SCS, is a popular falsehood; the graphs were not produced in the same study, and
NEITHER was from the Seven Countries Study.3,6,47
32
The Exclusion of France
Another popular criticism, similar to the first, is that Keys excluded countries that
didn’t fit his hypothesis. Namely, France was purportedly excluded due to the “French
Paradox.”
The “French Paradox” rests on the assumption that people in France tend to eat a diet
higher in saturated fat than people in other developed nations, yet experience lower rates
of heart disease.10 Critics believe that Keys was aware of this paradox and intentionally
left France out of the SCS. This, like the cherry-picking accusation, is demonstrably
incorrect.
France was not excluded
Missed by most revisionists is the fact that France was indeed invited to participate
in the Seven Countries Study. As previously mentioned, the SCS team wanted local
scientists from included countries to lead the project in their respective regions.3
Countries with willing, capable researchers, interest, and funds were welcome to join.
Some countries opted out due to lack of interest or resources; Sweden and Spain declined
for these reasons, respectively.8
Representatives from France had been included and Dr. Jacques Carlotti, a physician
from Paris, was part of the pilot study team in Nicotera, Italy.1,3,9 Ultimately,
representatives from France decided not to participate, possibly due to lack of desire, lack
of funding, or both. There was no explicit intent on the part of the American SCS
researchers to include or exclude France.
The “French Paradox” did not yet exist
The first reference to the now ubiquitous “French Paradox” in the literature was in
1981, making the concept a modern one. Pierre Ducimetière pointed out that the term was
33
first used in a paper by French epidemiologist J. L. Richard describing heart disease
incidence in French policemen.10 While it may have been noted that the French had rich
diets, it was not until the major diet cohort studies like Framingham and SCS were
published that the concept of France as a heart disease exception took hold.10 In fact, the
data available in the early and mid 1950s about the French diet did not suggest that
people of France had either an atypical diet or heart disease profile.6 Quite simply, even if
Keys and colleagues had been inclined to act on bias, there would have been no basis at
the time for a bias against the participation of France. All evidence regarding the SCS
suggests very much the contrary regarding Keys and colleagues, that they took pains and
went to great lengths to avoid bias in their methods.
The lack of French diet data
France was one of the many European countries occupied by Germany during World
War II, and thus experienced an extended period of dietary deprivation.7,51,53 As in much
of Europe, staple foods including animal foods and dairy were rationed and supplies
decreased throughout the war, in both Vichy and occupied France. Rationing was stopped
only in 1949 as France — like much of occupied Europe — struggled to recover after the
war’s end.51,52,59 Perhaps unsurprisingly, the food availability data for France in the late
1940s and early 1950s was not of the highest quality.
In fact, the Yerushalmy and Hilleboe paper of July 1957 inadvertently provides
evidence that international data for food availability and consumption in France were
poor.6 Of all 22 countries with varying qualities of dietary data, only France had no
available estimates for fats derived from animal and vegetable sources (proxies for
saturated and unsaturated fats).
The study record therefore belies the claim that Keys purposefully excluded France
from SCS due to concerns that France’s high saturated fat intake and low heart disease
rate would upset his hypothesis. There was little reliable dietary information available,
and none to foster any such inference. Further, the study’s design specifically included
34
countries with high fat intake and unknown levels of heart disease, such as Greece, in
order to test the relationship between total fat intake and coronary heart disease.3
Ultimately, SCS concluded that there was no observed relationship between the two; only
saturated fat showed a strong correlation.3
The French Paradox in the 1950s: Poor records and poor diagnoses
The scant evidence available in the early 1950s during the planning phase of SCS did
not show a French Paradox.
For example, see the 22-country chart from Yerushalmy and Hilleboe’s 1957 paper6
(page 23, page 35). France is represented by #8 on the chart. France does appear to have
extremely low heart disease rates, but it is also listed as having a total fat intake of less
than 30% of calories, an intake consistent with the often maligned initial dietary
guidelines of the United States.6 Few people who point to France to advocate the safety
of diets high in saturated fat encourage total fat intake this low.
Additionally, Yerushalmy and Hilleboe’s paper argued that the appearance of low
coronary heart disease mortality in France was a façade; due more to different diagnostic
criteria than to a truly lower mortality rate from heart disease.6 And so again, critics of
Keys from a half-century ago wind up refuting a popular claim among critics today.
In Keys’ 1953 review paper, he compared the dietary data of six countries using
mortality statistics from WHO Annual Epidemiological and Vital Statistics category “B-
26: Arteriosclerotic and degenerative heart disease.”6,47 Mortality from heart disease was
categorized as deaths per 100,000 people. In France, there were only 102 such deaths
ascribed to code B-26 from the 1951-1953 data, much lower than the United States’
number of 739. But Yerushalmy and Hilleboe point out that there is also the catch-all
heart disease category of “B-27: Other diseases of the heart.” In the United States, only
35 deaths per 100,000 were assigned to this category, while France had 180 per 100,000.6
If one were to combine deaths in France from B-26 and B-27, France would still have a
35
somewhat lower heart disease rate than the United States, but not impressively so, and by
no means “paradoxical.”
Below is the chart of 22 countries with France added back showing total deaths from
B-26 and B-27 included. The adjusted number is represented by the red dot.
Figure 3: Adapted from "Fat in the diet and mortality from heart disease: A methodologic note," 1957, New York State Journal of Medicine. Permission to reprint granted by the publisher.
If the deaths categorized as “B-27: Other diseases of the heart” are included in the
coronary heart disease mortality data, it is clear that France falls into line with an upward
slope demonstrating a correlation between fat intake and heart disease. Indeed, later
research has also brought to light the fact that inter-country variability in nationally
reported mortality data is a major weakness in ecological studies and may be responsible
for part of what we now think of as “The French Paradox.”10,19 Later epidemiological
36
studies on coronary heart disease across Europe showed that heart disease risk in France
is consistent with the geographic gradient seen in SCS that showed lower incidence in
southernmost regions and higher incidence moving north.60,61 Clear evidence describing
lower rates of coronary heart disease, but not all-cause mortality, in French men
disproportionate to what would be predicted based on traditional risk factors were not
defined until the 1980’s.62,63
SCS would go on to control for variability in cause-of-death diagnoses by using
study-specific researchers and standard and objective criteria to categorize and determine
official cause of death for the study data across all participating sites.3
In conclusion, the argument that The Seven Countries Study excluded France
due to knowledge of the French Paradox is false and anachronistic. The concept of
the French Paradox did not exist until decades after SCS was launched. Further, WHO
and FAO data available around the time of the start of SCS gave no reason for
researchers to think that the French ate a richer diet or experienced significantly less heart
disease than other similar countries in Europe.6 Finally, France was not excluded from
SCS at all, but was invited- and French researchers declined to participate.8
37
The Seven Countries Study: Dietary Data in Greece
Critics of SCS often cite concerns with methodology in addition to country selection.
One criticism, predominantly featured in the book Big Fat Surprise, was that one of the
dietary surveys conducted in Crete took place during Lent.11 Because Lent requires
fasting among adherent Orthodox Greek Catholics, the author concluded that saturated fat
and animal food intake were significantly underestimated in SCS. How would this have
impacted the findings? Following this argument, it’s possible that the Greeks were eating
a moderate or even high saturated fat diet that went undetected by researchers. If true,
this would weaken the argument that saturated fat intake was significantly associated
with incidence of and mortality from heart disease, since both Greek cohorts had an
extremely low incidence. However, this critique overlooks several key points.
First, the SCS researchers were aware some dietary recalls occurred during Lent.3,12
Statistical comparisons were done and there were no differences in intake between
Lenten and non-Lenten sampling periods. Sampling during Lent, rather than being a
researcher oversight, was a purposeful choice. If diet was significantly different during
this time, it was important for researchers to capture those data accurately.3,8
Additionally, the findings of SCS in Greece were consistent with an earlier and
unaffiliated dietary intake study in Greece, suggesting that Keys et al. did collect accurate
data.3,13,14
Dietary differences during Lent
There were two cohorts examined in Greece: Crete, and Corfu. These cohorts were
unique due to the extremely low levels of heart disease these participants experienced,
and also their high fat intake (around 40% of calories in Crete).1,3 Dietary intake was
measured via in-depth, seven-day, precisely weighed food records: two in Corfu and
three in Crete. To validate the findings, a representative subsample of each cohort was
chosen and followed for a week. Throughout this week, all food eaten by participants was
38
weighed and measured by researchers, and proportional physical samples of all food were
taken, freeze-dried, and shipped back to the lab for chemical analysis. Then “food tables”
based on seven-day weighed food records and chemical analyses were created.12
When all the data for dietary intake in Greece were compared, researchers found little
variation in total intake between different dietary surveys.3 The February survey in Crete
and part of the March-April survey in Corfu were taken during Lent. In a 1968
monograph containing all of the dietary survey and chemical analysis methodologies,
researchers made a specific note: “The Greek Orthodox rules are much more strict than
those of the Roman Catholic Church and the ‘fasting’ period of eight weeks should,
theoretically, have a major impact on the total diet for the year. However… actual
practice is different from Church prescription.” Total caloric intake was “substantially
unaffected,” and the proportion of energy intake from the macronutrients was similarly
unaffected.12
From the same paper, researchers broke down food intake by percent of energy from
specific foods and percent energy from macronutrients. The argument that including
Lent in the survey would lower the percent of calories coming from saturated fat —
often simplified as animal fat in dietary tables — is effectively moot. As shown below
in Figure 4 from the 1968 monograph, “Dietary studies and epidemiology of heart
diseases,” the same animal fat intake was observed during Lent in Corfu and a slightly
higher intake of animal fat during Lent in Crete.12
Similarly, it can be seen below in Figure 5, a table from the same paper, that there
was little difference in total macronutrient intake between seasons.12 Below, researchers
aligned dietary intake measured by seven-day surveys next to corresponding values found
by chemical analysis. Generally, chemical analysis slightly underestimates fat intake. In
the monograph, den Hartog et al. theorize that this was likely due to fat from foods
adhering to the sides of collection tubes.12 (Readers with any experience using bomb
calorimetry for high fat foods like fast food or snack chips have likely experienced this
frustrating phenomenon.)
39
Figure4:From “Dietary studies and epidemiology of heart disease,” 1968, Stichting tot wetenschappelijke Voorlichting op Voedingsgebied. Permission to reproduce granted by the publisher.
Figure5:From “Dietary studies and epidemiology of heart disease,” 1968, Stichting tot wetenschappelijke Voorlichting op Voedingsgebied. Permission to reproduce granted by the publisher.
Far from overlooking the problems inherent with collecting dietary data during
a time of religious fasting, researchers actively sought to see a difference.3,12 In the
Greek Orthodox church, adherent practitioners can have between 180-200 annual days of
religious fasting, a sum that makes up at least half the year.64 It would be irresponsible
not to include days of ritual fasting in populations with high adherence since this would
40
affect the average dietary intake of people over a calendar year, though adherence does
not appear particularly strict in this case.
Averaged, researchers determined that men in both Corfu and Crete consumed 7%
and 8% of their daily calories from saturated fat respectively, with the other fats coming
from monounsaturated fat, and very small amounts from polyunsaturated fat.3 Total fat
intake in Corfu was estimated at 33% and in Crete at 40%, which is consistent with the
values represented in Figure 5.12
The Rockefeller Foundation Study
Some concerns have been expressed that the dietary survey data aren’t accurate for
other reasons, such as a small subgroup sample size. If this were the case, data reliability
could be examined by comparing dietary data collected by SCS with the summary of data
from any other dietary studies or food availability data at the same time. Inaccuracy could
be suspected if the SCS findings were to show values significantly different from those of
other dietary surveys.
In her paper on the history of the Mediterranean diet concept, Marion Nestle pointed
out that in 1948, three years past ravages of World War II, the Greek government was
interested in improving health and quality of life for citizens. It therefore enlisted the help
of scientists and funding from the Rockefeller Foundation to measure lifestyle factors.14
Among other measurements, this study also did seven-day weighed dietary surveys
among a sample of Greek citizens in Crete. The study, led by epidemiologist Leland
Allbaugh, compared the results of their dietary survey with FAO availability data and
energy available based on household food stores.13
Data were collected in 1948 and results were published in 1953. The results are
shown below (concurrent U.S. food balance is included for comparison).13,14
41
Figure6From "Mediterranean diets: historical and research overview," 1995, American Journal of Clinical Nutrition. Permission to reproduce original figure granted by the author.
Though the table is organized differently, it is clear that the results, published 15
years before the 1968 monograph, largely agree with the data of SCS. Oils and fats are
29% of total energy while animal foods are 4% and dairy is 3%.13 Animal foods plus
dairy make up 7% of the 1953 findings, while the combination of animal protein and
animal fats (which are comprised of roughly the same foods) made up 8.1%. If any major
difference appears, it is that the later dietary survey of SCS finds Greek men consuming
slightly more total calories.12 This makes sense, considering that the first survey was
conducted only three years after WWII and only one in every six families reported having
satisfactory diets.13,14 Participants generally stated that they would prefer to have more
food, particularly meat and other animal foods.
There is little evidence that Greek citizens were eating higher amounts of saturated fat
and animal foods that were not picked up in either dietary survey.
42
In conclusion, insinuations that Greek dietary data in SCS were inaccurate, negligent,
or unrepresentative are unfounded. Sampling during Lent was purposeful in order to
ensure accurate average intakes in Crete and Corfu. Further, the dietary findings showed
that Greek citizens did not significantly alter dietary intake during Lent. These dietary
data are consistent with earlier, high quality, independent dietary surveys undertaken in
Crete in the late 1940s.
43
Sugar vs. Saturated Fat in Heart Disease
Multiple sources have reported that SCS neglected deleterious effects of sugar on
health in favor of promoting a low-fat agenda.11,15 One accusation is that the regression
analyses used in SCS did not adjust for sugar as a possible confounder.65 Still another
stated that a later 1999 reanalysis of SCS data found that sugar was more highly
correlated with heart disease than saturated fat, proving that sugar, and not saturated fat,
was a culprit.11,15
Like the other arguments addressed in this paper, these criticisms do not withstand
scrutiny with attention to primary sources and historical documents. Sugar was adjusted
for in the multivariable analysis, as was discussed in the findings section above.3 The
1999 reanalysis did not upend the findings of the 1980 SCS publication; in fact, findings
remained consistent.
Adjustingforsugar
Summaries are quick to state that SCS did not consider sugar when statistical
analyses were performed searching for relationships between diet and heart disease. Had
these correlations been done, critics attest, it would be clear that sugar correlates as well
as or better than saturated fat with the incidence of heart disease.
The SCS investigators found and reported that sugar did correlate with heart disease.3
However, Keys et al. did not find that sugar was significantly related to heart disease
when the model was adjusted for saturated fat. This was consistent with experimental
findings at the time that failed to show a physiological mechanism directly connecting
sugar intake and heart disease.66 Ultimately, sugar received only a few paragraphs in the
1980 SCS publication not because there was something to hide, but because there was
nothing to show. Sugar was analyzed in the same way as saturated fat, and the findings
that resulted were the findings reported.3
44
Mechanistic involvement in heart disease
John Yudkin, a professor of nutrition in London, posited in the 1960s that heart
disease could be related to hyperinsulinemia caused by a high sugar diet.66 In his 1971
paper, “Sugar Intake and Coronary Heart Disease,” A.R.P. Walker explained the history
and state of nutrition science regarding sugar intake and heart disease. He included
descriptions of Yudkin’s human trials, which were published more than 12 years after the
start of SCS.66 During the four-week trials, participants consumed up to 40% of their
daily calories from sucrose for two weeks. Findings showed that the high sugar diet
caused no differences in glucose tolerance, serum cholesterol, or serum phospholipid
levels.66
However, Walker noted that one-third of subjects on the high sugar diet exhibited
hyperinsulinemia, and patients with previously established peripheral vascular disease on
the high sugar diets also experienced this outcome.66 This led to the suggestion that the
high sugar diet could cause adverse effects in people with a predisposition to develop
“sucrose-induced hyperinsulinism,” as Yudkin called it. These trials, while intriguing,
suffered from a confounding factor: The diets were not isocaloric. Those experiencing the
hyperinsulinism were gaining weight. Walker reported that in subsequent trials, Mann et
al. tested the results of reducing sugar intake in participants, and weight loss was a
confounding factor.66 As has been seen numerous times in human diet trials, it is difficult
to adjust intake of a single nutrient or macronutrient while maintaining isocaloric intake.
Further, the majority of case control and population observational studies that had
been completed by the mid-20th century found that those with coronary heart disease
weren’t consuming significantly more sugar than those without it.66 Dietary studies
swapping sucrose for other types of carbohydrate — namely starch — also found that
sugar had no overall relationship to health, although some studies found slightly raised
plasma triglycerides. In some case-control studies, the amount of sugar consumed by
those affected by heart disease was much less than the amount consumed by the control
45
groups without heart disease.66 This was difficult to square with a risk of heart disease
proportional to sugar intake.
In addition to Yudkin’s lack of statistical calculations — his association between
population data on sugar intake and heart disease relied on a scatter plot and no
calculated correlation coefficients — his population-wide observational studies suffered
from selection bias (a fault frequently attributed to Keys). In a critique by Ancel Keys, he
pointed out that countries with high sugar intakes, but low reported heart disease, like
Brazil, Colombia, Nicaragua, Uruguay, and Mauritius, were left out of Yudkin’s
analysis.38 It didn’t help that other scientists had found that sugar intake was statistically
correlated with a smoking habit, further muddying the waters in the crude ecologic
relationship between sucrose and heart disease.66
To much of the nutrition community at the time, links between sugar intake and heart
disease risk were simply unconvincing, especially when compared to the amount of data
linking saturated fat and heart disease.66
Figure7:From "Sugar intake and coronary heart disease," 1971, Atherosclerosis. This image has been reproduced from widely available reproductions accessible online. Permission to reproduce the original has been sought from the publisher, and a response is pending at this time.
46
A follow-up in 1999
A recent lay press article that dealt with the SCS — scintillatingly titled “The Sugar
Conspiracy” — tried to make the case that SCS was later discredited by one of its own
lead authors, Dr. Alessandro Menotti.15
“Years later,” the article said, “the Seven Countries study’s lead Italian researcher,
Alessandro Menotti, went back to the data, and found that the food that correlated most
closely with deaths from heart disease was not saturated fat, but sugar.”15
The correlations and conclusions mentioned above are not reported in the 1999
paper.16 When asked for comment, Dr. Menotti could not account for this conclusion: “I
never said that sugar was more correlated to coronary heart disease than other food
groups.”67
The 1999 follow-up on the dietary data was an attempt to look at the relationship
between specific foods and heart disease mortality. In earlier iterations of the study, each
food was broken down and analyzed according to its macronutrient components; this new
study took a “whole foods” approach.16
Unlike the conclusion reported in “The Sugar Conspiracy,” this analysis did not find
that sugar was more associated than saturated fat with heart disease mortality. Saturated
fat was not assessed individually, since it is a nutrient component and not a food.16
However, foods that are often considered proxies for saturated fat, like butter and other
animal fats, were assessed.
A significant positive association was found between sugar and death from coronary
heart disease (r=+0.60) that was consistent with earlier findings of the SCS.3,16 Foods that
contained high amounts of saturated fat, however, such as butter (r=+0.887), pastries (a
mix of sugar, carbohydrates, and fat (+0.752)), and all animal foods (r=+0.798), all had
stronger correlations with mortality from heart disease than did sugar.16
47
Accordingly, it is false to contend that SCS ignored or didn’t address the possible
relationship of sugar to heart disease. Today, there is more evidence that sugar and other
refined carbohydrates are not an advisable replacement for saturated fat sources in a
healthy diet.39,68–70 At the time of SCS, no experimentally determined mechanism existed
that linked sugar intake to heart disease. More importantly, sugar was not ignored.
Researchers assessed sugar intake, but no correlation between sugar and coronary heart
disease was found once saturated fat intake was controlled for in the analysis.3 In other
words, sugar intake in the SCS analysis correlated with coronary heart disease only to the
extent it correlated with intake of saturated fat.
48
STRENGTHS AND LIMITATIONS OF THE SEVEN COUNTRIES STUDY
At the time of its inception, SCS was groundbreaking. Comparing incidence rates of
coronary heart disease across populations with vastly different traditional diets provided
information on risk that was not previously available.3 Though definitive causes of heart
disease could not be determined, it was clear that coronary heart disease was not simply a
matter of fate, but could be influenced by environment, lifestyle, and diet.
Prior to SCS, Keys and other researchers had noted seemingly different incidences of
chronic diseases, or diseases of aging, among populations they studied. Observational
studies of individual cultures limited researchers’ ability to compare diet, lifestyle, and
risk factors across cultures due to differences in study design, methodological quality,
and measured outcomes. Previously, national mortality and dietary statistics were highly
variable, which complicated attempts to compare diet and disease prevalence between
nations, such as in Keys’ 1953 paper and Yerushalmy and Hilleboe’s subsequent
response.6,7,47,55 SCS was able to bridge the gap by deliberately choosing vastly different
cohorts and comparing each with strictly standardized measures.
The development of these methods, including the Minnesota Code, which
standardized methods for measuring incidence of heart disease and confirming heart
disease mortality, represented a major breakthrough in epidemiology.21 Dietary records
were equally impressive and much more thorough than many recall tools used today, and
the chemical characterization of nutrient intake for the regions was unique.3,12
But like all scientific studies, especially groundbreaking ones, the study had
limitations, which were largely acknowledged and dealt with in SCS publications.
SCS was unable in its earlier follow-up years to detect any significant excess risk for
coronary heart disease or mortality associated with overweight or obesity.3 In fact, the
1980 study report for the 10-year coronary rates found that BMI was inversely correlated
49
with death rate in all cohorts. At the time, most participants were at least moderately
active, healthy men in their 40s and 50s, and the overall prevalence of obesity was low
compared to today’s levels.3 It is possible that SCS suffered from the same issue here that
occurred with smoking — there was not enough variation in individual or population
rates to reveal a meaningful correlation. It is consistent with modern research to find an
increased risk of death at weights below a BMI of 18, which was common in this study.
However, the average BMI in most countries hovered around 25, without extreme
variation within or among regions — meaning that most people were around normal
weight or slightly overweight.3 This would prevent SCS from finding many effects of
obesity or overweight.
Secondly, due to the lack of physiological evidence or plausible mechanism
connecting sucrose and atherosclerosis, refined sugar and carbohydrates were not given
prime attention. Modern epidemiological studies modeling the effects of swapping
saturated fat for refined carbohydrates and sugar suggest that there is no change in
risk when one is substituted for the other.71 In other words, the models suggest that
butter should not be swapped for sugar, or vice versa, to avoid excess heart disease risk.
Numerous studies observe that sources of polyunsaturated fats and whole grain or
unrefined carbohydrate products are superior.39,40 It is possible that this relationship was
not seen in SCS because sugar and saturated fat intake were highly correlated.3 Even so,
saturated fat was independently associated with heart disease rates, while sugar was not.
The SCS team was unable to take all of the measurements they wanted due to
practical constraints. For example, they initially were interested in getting glucose
tolerance tests and fractionated lipoprotein cholesterol, but were limited by the difficulty
of collecting these measures, shipping them to the central laboratory, and the associated
expense. Glucose tolerance would have put a major time burden on the participants
coming in and may have hurt participation rates. Additionally, lack of funds for incidence
studies in the U.S. cohort after the five-year mark meant that researchers could assess
outcomes only in coronary heart disease mortality in American men during the later years
of the study.3
50
The ecological design of SCS imposed important limitations, probably most
importantly difficulty in controlling for confounding. Because the primary outcomes
compared cohorts rather than individual participants’ values, correlations were drawn
using only 16 data points; the cohort means for measurements taken. This limited how
much could be inferred from more complex statistics employed in the study, like the
multivariable regression, because findings are limited by a modest sample size and
degrees of freedom. Additionally, sites were non-random and not meant to be
representative samples of each country, but purposefully varied in diet.18 This allowed
Keys et al. to see stark contrasts between diet, lifestyle, and disease outcomes but made
them less able to infer that these same variations in risk would carry over to populations
outside of the study samples.3
Though the 1980 publication gave little attention to foods, including sugar, BMI, or
smoking, in relation to total mortality, these are all limitations caused by lack of
statistical power or logistical challenges. Even with perfect data collection and analysis,
SCS can only report the comparative risks related to diet and heart disease within the
specific populations it studied.3 The diets of all cohorts in 1959-1964, when SCS dietary
data collection took place, were dissimilar to diets we think of as the “Standard American
Diet” today. Populations were not eating the high calorie, highly processed foods now so
common in diets across the globe, so it is sensible to be cautious when extrapolating the
findings of this single study.
51
CONCLUSIONS
SCS provided invaluable information about the connections between diet and
coronary heart disease, in addition to developing important tools for standardizing
nutrition and diet research. Monographs published with five and 10-year follow-up data
showed that coronary heart disease and mortality were significantly associated with diets
higher in saturated fat.2,3 Follow-up data from 15 years of tracking all-cause and coronary
heart disease mortality found that diets high in saturated fats were associated with
increased risk for dying of heart disease, consistent with earlier findings.4 Allegations that
SCS was deliberately designed in order to find specific, preconceived outcomes are
decisively without merit.
Like all scientific studies, SCS had limitations, largely due to its ecological design,
and should be interpreted in the context of other evidence. Other concurrent, carefully
conducted studies, including the Framingham Heart Study, reached similar conclusions
for the associations between blood cholesterol levels and coronary heart disease risk.24
These other studies then inspired a series of randomized controlled dietary trials testing
the effects of fat substitution on heart disease risk that had findings consistent with those
of the large-scale epidemiological nutrition studies.72 The aggregation of this evidence,
together with other scientific findings and many other influences, contributed to the
original 1980 dietary guidelines and still influences the guidelines today.73 All sets of
dietary guidelines have advocated for limited intake of saturated fats as well as
sugars.73,74 Of note, Ancel Keys was only briefly interested in total fat intake as an
important variable; his own data quickly dissuaded him from that focus, and shifted his
attention to saturated fat.
Misleading, or negligent criticism of seminal research in nutrition science undermines
the credibility of all science and the process by which understanding advances. Detractors
invite false equivalence by allowing studies of smaller impact and dubious quality to be
compared against large-scale, scrupulously conducted research. Current examples of the
insidious dangers of such a process include climate change denial, and false allegations
52
about the effects of childhood immunization. As a result of the latter, the United States
and other developed nations have seen an increase in vaccine preventable diseases and
subsequently an increase in deaths from these diseases.75 The public health impact of
false or misleading narratives about nutrition research is potentially much larger even
than that of vaccines.
Heart disease is the number one killer in America and worldwide.76,77 The thesis of
the many articles and books disparaging SCS and the life work of Ancel Keys is that
serum cholesterol and saturated fat foods are not related to heart disease. Heart disease
has decreased significantly since the 1950s, however, and despite the increase in obesity
during that same time period. These declines are thought to be due to public health
interventions addressing lifestyle practices, including a significant shift from dietary
sources of saturated fats to sources of unsaturated fats, as well as medical advances.76 In
North Karelia, Finland, where the SCS findings and the work of Ancel Keys were used
directly and with fidelity as the basis for intervention at the community level, heart
disease rates plummeted over 80% in the decades that followed, and average life
expectancy increased by a full ten years.78–80 There, too, along with other changes,
calories from dietary sources of saturated fat, especially dairy fat, were replaced largely
with dietary sources of unsaturated fats.
Recently, the decline in coronary heart disease incidence in the United States has
slowed.76 Rates of heart disease in North Karelia, Finland, have risen slightly for the first
time in half a century, in apparent tandem, whether coincidentally or causally, with
popular messaging about the advantages of eating more “meat, butter, and cheese.”81
In the preface for the 2002 monograph Prevention of coronary heart disease: Diet,
lifestyle and risk factors in the Seven Countries Study, Ancel Keys summarizes:
“The results of the Seven Countries Study… should be viewed in this broader context
of the role of the diet, lifestyle and risk factors in relation to all causes of deaths and to
longevity. This should be the ultimate goal of epidemiologic research.”18
53
Like all studies, SCS should be analyzed and utilized in context, with strengths and
weaknesses acknowledged. By deliberately or carelessly misrepresenting historical
events, distorting scientific findings, and misstating researcher intent, modern critics of
the SCS routinely impede, rather than advance, understanding in nutritional
epidemiology.
54
EDITORIAL EPILOGUE: The Legacy of Ancel Keys and the SCS By David Katz, MD, MPH
This White Paper, relying preferentially and to the extent possible on primary sources
and first-hand accounts, decisively falsifies the popular disparagements directed at Ancel
Keys and the Seven Countries Study. The Seven Countries Study included exactly seven,
and neither six nor 22, countries. Keys and colleagues did not cherry-pick the
participating countries; they did not exclude France; they did not present or graph their
data selectively; they did include dietary intake surveys in Greece during Lent
intentionally, for reasons clearly articulated at the time, and with proof that this did not
introduce any distortions; and they did analyze sugar in all the same ways they analyzed
saturated fat, and reported just what they found.
The popularization of flagrant falsehoods about all of the above is possible for several
reasons: the primary source materials are now decades old, and few have or bother to
seek recourse to them; the Internet lacks reliable editorial filtering or fact-checking
mechanisms, so idle opinion passes routinely for expertise; repetition in cyberspace and
social media is nearly effortless, and oft-repeated falsehoods can “drown out” evidence
and truth with sheer volume; messages are invariably distorted in sequential transmission,
as in the party game, “telephone;” and, conveniently for his detractors, Keys is
unavailable to defend himself posthumously against even the most readily refuted
falsehoods.82,83
This much the paper establishes. It stops there, however, because that was the charge
to which it was responsive: a reality check about history, and a response to revisionist
alternatives. But what about the legacy of Keys and the SCS and its reverberations into
modern public health?
Keys has been routinely implicated in the low-fat dietary digression that brought us
such low-fat junk foods as Snackwell cookies; that increased our intake of refined grains,
added sugar, and total calories; and that failed to advance leading public health objectives
55
related to diet.84,85 This charge, too, is clearly false. Primary sources, and first-hand
accounts by investigators both indicate that Keys observed an association between (total)
dietary fat intake and heart disease in his early data. According to colleagues, however,
Keys routinely did with such observations what good scientists should do: turned them
into testable hypotheses, not immutable convictions.49 He proceeded to test this
hypothesis himself, and concluded long before the dawn of the “low fat” diet era that
total dietary fat quantity was unimportant, while the sources and quality of fat were
important. Over the course of his career, he advocated for restricting saturated (and,
albeit with lesser attention, trans fat), fat by shifting from animal-food-centric to plant-
food-centric diets; and lent his strongest support to the natively high-unsaturated-fat
Mediterranean diet. Keys published a Mediterranean diet cookbook in 1959.86
Misadventures in low-fat eating, to say nothing of low-fat junk foods, cannot
legitimately be attributed to any position espoused by Keys over the course of his career.
That there was a foray into misguided applications of a low-fat dietary approach with
ramifications still relevant today is undeniable. Reasons for it are complex, contentious,
and beyond the scope of this paper- and perhaps fodder for another. But it is clear that
responsibility cannot legitimately be assigned either to Keys, or the Seven Countries
Study.
There are published claims that Keys was unduly confident about specific
associations, such as that between saturated fat and heart disease, given the ecological
nature of the SCS. However, the written record and other first-hand accounts indicate
that (a) Keys almost without fail carefully avoided any temptation to overstate the
significance of his findings in published work; and (b) he based his
impressions/conclusions on the overall mass of evidence, not just the SCS. Importantly,
his views evolved over the course of his career in accord with the evolving evidence base.
Among the currently popular narratives is that sugar is the culprit saturated fat was
long thought to be; that the contributions of sugar to cardiovascular disease were willfully
concealed by vested interests; and that Keys and the SCS were complicit in this. The
56
reality here is that efforts to veil the harms of sugar obviously failed, since the Dietary
Guidelines for Americans have emphasized limiting sugar intake since the first in
1980.73,87 Keys, and the SCS, gave sugar its day in court, and reported just what they
found. Studies conducted by Keys and his SCS colleagues did reveal an association
between sugar intake and cardiovascular disease, albeit a weaker one than for dietary
sources of saturated fat. Nothing in this body of work ever encouraged or justified the
substitution of added sugar for saturated fat, and Keys was never an advocate for any
such thing. There is more than one way to eat badly, and if the American public has been
committed to exploring them sequentially, blame for it cannot be laid at the door- or now
the grave- of Ancel Keys.
Viewing serum lipids as a key mediator of the effects of diet on the cardiovascular
system, Keys was initially interested in total cholesterol, and then both LDL and HDL,
but settled on the observation that LDL was of singular importance. His view on this
matter seems to have anticipated by some decades the very conclusion now favored by a
consensus of experts, and the shifting weight of relevant evidence.88,89
The principal conclusion of the SCS germane to public health practice was that heart
disease was preventable with lifestyle and not an inevitable consequence of aging (a new
concept at the time), and that dietary intake of saturated fat from its prevailing food
sources (meat, processed meat, and dairy products then; with fast food and various
processed foods appended now) should be reduced and replaced by whole foods, mostly
plants, and the unsaturated oils derived from olives in particular, but also nuts, seeds,
avocados, fish, and seafood- to lower cardiovascular disease risk. That conclusion has
since been reached by diverse experts in diverse settings on the basis of diverse research
all independent of the SCS.40,71,74,90–104
That conclusion was converted into practice in North Karelia, Finland, with fidelity to
the findings of the SCS rather than the distortions that produced low-fat junk food in the
United States. The result was a reduction in incident coronary disease of more than 80%,
and an average addition to life expectancy of more than ten years.80 While these results
57
were not solely attributable to a shift from animal foods to plant foods, with an attendant
drop in saturated fat intake, that was a prominent component of the campaign.78
Keys may have been somewhat dismissive of trans fat, but mostly because it occupied
a very tiny niche in the American diet at the time of his initial work. He may have been
somewhat dismissive of the importance of obesity to cardiovascular outcomes, but mostly
because weight varied rather little among the workers enrolled in the SCS. He may have
been somewhat inattentive to the adversities of smoking, again for lack of variation in the
exposure among his study participants. Overall, though, based on both data he generated,
and data generated by others, Keys evolved positions remarkably consistent with those
reached independently on the basis of the most current evidence on: overall dietary
pattern; total dietary fat; saturated fat; unsaturated fat; sugar; LDL; and dietary
cholesterol, which Keys recognized as relatively unimportant to serum cholesterol levels.
None of the first-hand accounts of Keys included the claim that he never made
mistakes, or the recommendation that he be canonized. All concurred, however, that he
was a diligent, meticulous, groundbreaking scientist who followed the data where they
led. On the basis of data, and perhaps good intuition as well, Keys anticipated the
evidence and consensus-based positions of public health nutrition in 2017 with
extraordinary accuracy and consistency. When public health nutrition in the modern era
has gone awry, it has never done so in accord with the findings and positions of Ancel
Keys. It has done so because such findings have been distorted; such positions
misrepresented; and the important lessons of this period of nutrition history, and the
singular contributions of Ancel Keys, forgotten and replaced with false narratives.
-David L. Katz, MD, MPH
Founder & President, The True Health Initiative
Immediate Past President, The American College of Lifestyle Medicine
Director, Yale University Prevention Research Center
Griffin Hospital
Derby, CT
58
REFERENCES 1. Keys,A.,Aravanis,C.,Blackburn,H.W.,vanBuchem,S.P.,Buzina,R.,
Djordjevic,B.S.,Dontas,A.S.,Fidanza,F.,Karvonen,M.J.,Kimura,N.,Lekos,D.,Monti,M.,Puddu,V.TaylorHL.Epidemiologicalstudiesrelatedtocoronaryheartdisease:Characteristicsofmenaged40-59insevencountries.ActaMedScand.1966;460:1-392.
2. KeysA.Coronaryheartdiseaseinsevencountries.Circulation.1970;41(S1):118-139.
3. Keys,A,Aravanis,C.,Blackburn,H.,Buzina,R.,Djordjevic,B.S.,Dontas,A.S.,Fidanza,F.,Karvonen,M.J.,Kimura,N.,Menotti,A.,Mohacek,I.,Nedeljkovic,S.,Pussu,V.,Punsar,S.,Taylor,H.L.,VanBuchemFSP.SevenCountries:AMultivariateAnalysisofDeathandCoronaryHeartDisease.Cambridge,MA:HarvardUniversityPress;1980.
4. KeysA,MenottiA,KarvonenMJ,etal.Thedietand15-yeardeathrateinthesevencountriesstudy.AmJEpidemiol.1986;124(6):903-915.
5. FraminghamHeartStudy.ResearchMilestones.https://www.framinghamheartstudy.org/about-fhs/research-milestones.php.Published2017.
6. YerushalmyJ,HilleboeH.FatintheDietandMortalityfromHeartDisease:AMethodologicNote.NYStateJMed.1957:2343-2354.
7. KeysA.Epidemiologicaspectsofcoronaryarterydisease.JChronicDis.1957;6(5):552-559.http://www.ncbi.nlm.nih.gov/pubmed/13475453.
8. BlackburnH.personalcommunication.2017.9. BlackburnH,KromhoutD,AM.TheSevenCountriesStudy:Ascientific
adventureincardiovasculardiseaseepidemiology.1994:15-37-159.10. DeLeirisJ,BoucherF,DucimetiereP,HoldsworthM.TheFrenchParadox :
FactorFiction ?ExpertAnswerstoThreeKeyQuestions.DialoguesCardiovascMed.2008;13(3).
11. TeicholzN.TheBigFatSurprise:WhyButter,Meat,andCheeseBelonginaHealthyDiet.NewYork:Simon&Schuster;2014.
12. denHartogC,BuzinaK,FidanzaF,KeysA,RoineP.DietaryStudiesandEpidemiologyofHeartDiseases.TheHague;1968.http://edepot.wur.nl/380320.
13. AllbaughLG.Crete:ACaseStudyofanUnderdevelopedArea.PrincetonUniversityPress;1953.
14. NestleM.Mediterraneandiets:historicalandresearchoverview.AmJClinNutr.1995;61(suppl):1313S-1320S.
15. LeslieI.Thesugarconspiracy.Guard.2016.https://www.theguardian.com/society/2016/apr/07/the-sugar-conspiracy-robert-lustig-john-yudkin.
16. Menottia,KromhoutD,BlackburnH,FidanzaF,BuzinaR,Nissinena.Foodintakepatternsand25-yearmortalityfromcoronaryheartdisease:cross-culturalcorrelationsintheSevenCountriesStudy.TheSevenCountriesStudyResearchGroup.EurJEpidemiol.1999;15(6):507-515.
59
doi:10.1023/A:1007529206050.17. KeysA,MenottiA,AravanisC,etal.TheSevenCountriesStudy:2,289Deaths
in15Years.1984;154:141-154.18. KromhoutD,MenottiA,BlackburnH.PreventionofCoronaryHeartDisease:
Diet,LifestyleandRiskFactorsintheSevenCountriesStudy.NewYork:KluwerAcademicPublishers;2002.doi:10.1007/978-1-4615-1117-5.
19. LozanoR,MurrayC,LopezA.Miscodingandmisclassificationofischaemicheartdiseasemortality.WorldHealth.2001;12(12):1-19.
20. RoseG,BlackburnH.CardiovascularSurveyMethods.WorldHealthOrganizationTechnicalReportSeriesNo.56.Geneva:WorldHealthOrganization;1968.
21. BlackburnH.Electrocardiographicclassifictionforpopulationcomparisons:TheMinnesotacode.JElectrocardiol.1969;2(1):5-9.doi:https://doi.org/10.1016/S0022-0736(69)80044-0.
22. BlackburnH,KeysA,SimonsonE,RautaharjuP,PuntsarS.TheElectrocardiograminPopulationStudiesAClassificationSystem.Circulation.1960;XXI:1160-1175.doi:https://doi.org/10.1161/01.CIR.21.6.1160.
23. KromhoutD,KeysA,AravanisC,etal.Foodconsumptionpatternsinthe1960sinsevencountries.AmJClinNutr.1989;49(5):889-894.http://ajcn.nutrition.org/content/49/5/889.abstract.
24. MahmoodaSS,LevyD,VasanRS,WangTJ.TheFraminghamHeartStudyandtheepidemiologyofcardiovasculardiseases:Ahistoricalperspective.Lancet.2014;383(9921):1933-1945.doi:10.1016/S0140-6736(13)61752-3.The.
25. TsaoCW,VasanRS.CohortProfile:TheFraminghamHeartStudy(FHS):Overviewofmilestonesincardiovascularepidemiology.IntJEpidemiol.2015;44(6):1800-1813.doi:10.1093/ije/dyv337.
26. KeysA,TaylorH,BlackburnH,BrozekJ,AndersonJ,SimonsonE.CoronaryHeartDiseaseamongMinnesotaBusinessandProfessionalMenFollowedFifteenYears.Circulation.1963;28(3):381-395.doi:10.1161/01.CIR.28.3.381.
27. BorhaniNO,HechterHH,BreslowL.ReportofaTen-YearFollow-UpStudyoftheSanFranciscoLongshoremen:MortalityFromCoronaryHeartDiseaseandFromAllCauses.JChronicDis.1963;16:1251-1266.http://www.ncbi.nlm.nih.gov/pubmed/14104335.
28. BeatonGH,MilnerJ,CoreyP,McGuireV,Al.E.Sourcesofvariancein24-hourdietaryrecalldata:Implicationsfornutritionstudydesignandinterpretation.AmJClinNutr.1979;32(12):3546-3559.
29. JacobsDR,AndersonJT,BlackburnH.Dietandserumcholesterol:Dozerocorrelationsnegatetherelationship?AmJEpidemiol.1979;110(1):77-87.doi:10.1093/oxfordjournals.aje.a112791.
30. LiuK,StamlerJ,DyerA,McKeeverJ,McKeeverP.Statisticalmethodstoassessandminimizetheroleofintra-individualvariabilityinobscuringtherelationshipbetweendietarylipidsandserumcholesterol.JChronicDis.1977;31(6-7):399-418.doi:https://doi.org/10.1016/0021-9681(78)90004-8.
31. KaplanH,ThompsonRC,TrumbleBC,etal.CoronaryatherosclerosisinindigenousSouthAmericanTsimane:Across-sectionalcohortstudy.TheLancet.2017.
60
32. A.MJ,MasesaZ,StrømmeSB,AlE.DailyenergyexpenditureandcardiovascularriskinruralandurbanBantuTanzanians.BrJSportsMed.2010;44:121-126.
33. OrlichMJ,FraserGE.VegetariandietsintheAdventistHealthStudy2 :areviewofinitial.AmJClinNutr.2014;100:2-7.doi:10.3945/ajcn.113.071233.Am.
34. BlackburnH.20th-Century“MedicalMarcoPolos”intheOriginsofPreventiveCardiologyandCardiovascularDiseaseEpidemiology.AmJCardiol.2012;109(5):756-767.doi:10.1016/j.amjcard.2011.10.038.
35. KromhoutD.PersonalCommunication.2017.36. BlackburnH.PersonalCommunications.2017.37. KatzD,WildD,ElmoreJ,LucanS.Jekel’sEpidemiology,Biostatistics,Preventive
Medicine,andPublicHealth.4thed.Saunders;2013.38. KeysA.Sucroseinthedietandcoronaryheartdisease.Atherosclerosis.
1971;14(2):193-202.doi:10.1016/0021-9150(71)90049-9.39. HuF,WillettW.Optimaldietsforpreventionofcoronaryheartdisease.JAm
MedAssoc.2002;288(20):2569-2578.http://dx.doi.org/10.1001/jama.288.20.2569.
40. KatzDL,MellerS.CanWeSayWhatDietIsBestforHealth?AnnuRevPublicHealth.2014;35(1):83-103.doi:10.1146/annurev-publhealth-032013-182351.
41. KeysA,AndersonJT,GrandeF.Predictionofserum-cholesterolresponsesofmantochangesinfatsinthediet.Lancet.1957;273(7003):959-966.
42. GrandeF,AndersonJT,KeysA.Theinfluenceofchainlengthofthesaturatedfattyacidsontheireffectonserumcholesterolconcentrationinman.JNutr.1961;74:420-428.
43. BorénJ,WilliamsKJ.Thecentralroleofarterialretentionofcholesterol-richapolipoprotein-B-containinglipoproteinsinthepathogenesisofatherosclerosis.CurrOpinLipidol.2016;27(5):473-483.doi:10.1097/MOL.0000000000000330.
44. @ziobrando.Thebullshit[asymmetry].2013.https://twitter.com/ziobrando/status/289635060758507521.
45. SevenCountriesStudy.Wikipedia.46. KeysA.theDiet/HeartControversy.Lancet.1979;314(8147):844-845.
doi:10.1016/S0140-6736(79)90331-3.47. KeysA.Atherosclerosis:aprobleminnewerpublichealth.JMtSinaiHospNY.
1953;20(2):118-139.48. EstruchR,RosE,Salas-SalvadóJ,etal.PrimaryPreventionofCardiovascular
DiseasewithaMediterraneanDiet.NEnglJMed.2013;368:1279-1290.doi:10.1056/NEJMoa1200303.
49. KromhoutD,KatanMB,BlackburnH,MenottiA,JacobsDR,WillettWC.PersonalCommunication.2017.
50. JacobsDR.PersonalCommunication.2017.51. KesternichI,SiflingerB,SmithJP,WinterJK.TheEffectsofWorldWarIIon
EconomicandHealthOutcomesacrossEurope.RevEconStat.2014;96(1):103-118.doi:10.1162/REST_a_00353.
61
52. LoweK.SavageContinent:EuropeintheAftermathofWorldWarII.NewYork:St.Martin’sPress;2012.
53. CollinghamL.TheTasteofWar:WorldWarIIandtheBattleforFood.NewYork:ThePenguinPress;2012.
54. BlackburnH.FamousPolemicsonDiet-HeartTheory.http://www.epi.umn.edu/cvdepi/essay/famous-polemics-on-diet-heart-theory/%0D%0D.
55. HilleboeH.SomeEpidemiologicAspectsofCoronaryArteryDisease.JChronicDis.1957;6(3):210-228.doi:https://doi.org/10.1016/0021-9681(57)90003-6.
56. Bradford-HillA.TheEnviromentandDisease:AssociationorCausation?ProcRSocMed.1965;58:295-300.doi:DOI:10.1016/j.tourman.2009.12.005.
57. StanhopeB-J,BurdetteWJ,CochranWG,etal.ReportoftheAdvisoryCommitteetotheSurgeonGeneralofthePublicHealthService.…,EducWelfare,PublicHealServ….1964:1-386.http://profiles.nlm.nih.gov/NN/B/B/M/Q/_/nnbbmq.ocr.
58. BlackburnH,LabartheD.Storiesfromtheevolutionofguidelinesforcausalinferenceinepidemiologicassociations:1953-1965.AmJEpidemiol.2012;176(12):1071-1077.doi:10.1093/aje/kws374.
59. DavisM.HowWorldWarIIshapedmodernFrance.Euronews.May2015.http://www.euronews.com/2015/05/05/how-world-war-ii-shaped-modern-france.
60. KeysA.Wine,garlic,andCHDinSevenCountries.Lancet.1980;315(8160):145-146.doi:https://doi.org/10.1016/S0140-6736(80)90620-0.
61. Tunstall-PedoeH,KuulasmaaK,MarkkuM,TolonenH,RuokokoskiE,AmouyelP.Contributionoftrendsinsurvivalandcoronary-eventratestochangesincoronaryheartdiseasemortality-10-yearresultsfrom37WHOMONICAProjectpopulations.Lancet.1999;353(9164):1547-1557.doi:https://doi.org/10.1016/S0140-6736(99)04021-0.
62. DucimetiereP,RichardJL,CambienF,RakotovaoR,ClaudeJR.Coronaryheartdiseaseinmiddle-agedFrenchmen:ComparisonsbetweenParisProspectiveStudy,SevenCountriesStudy,andPoolingProject.1980;315(8182):1346-1350.doi:https://doi.org/10.1016/S0140-6736(80)91796-1.
63. Tunstall-pedoeH.Autrespays,Autresmoeurs.TheoriesonwhytheFrenchhavelessheartdiseasethantheBritishReinvestigateandchangethetreatment.BrMedJ.1988;297(6683):1559-1560.
64. SarriKO,LinardakisMK,BervanakiFN,TzanakisNE,KafatosAG.GreekOrthodoxfastingrituals:ahiddencharacteristicoftheMediterraneandietofCrete.BrJNutr.2004;92(2):277.doi:10.1079/BJN20041197.
65. LustigR.FatChance:BeatingtheOddsAgainstSugar,ProcessedFood,Obesity,andDisease.NewYork:Penguin;2013.
66. WalkerARP.Sugarintakeandcoronaryheartdisease.Atherosclerosis.1971;14:137-152.internet:http://edepot.wur.nl/380320.
67. MenottiA.Personalcommunication.2017.68. MensinkRP,ZockPL,KesterADM,KatanMB.Effectsofdietaryfattyacidsand
62
carbohydratesonserumtotaltoHDLcholesterolandserumlipidsandapolipoproteins:ameta-analysisof60controlledtrials.2003;77:1146-1155.
69. Siri-tarinoPW,SunQ,HuFB,KraussRM.Saturatedfat,carbohydrate,andcardiovasculardisease.AmJClinincalNutr.2010;(5):502-509.doi:10.3945/ajcn.2008.26285.INTRODUCTION.
70. WillettWC.Dietaryfatsandcoronaryheartdisease.JInternMed.2012;272(1):13-24.doi:10.1111/j.1365-2796.2012.02553.x.
71. LiY,HrubyA,BernsteinAM,etal.SaturatedFatasComparedWithUnsaturatedFatsandSourcesofCarbohydratesinRelationtoRiskofCoronaryHeartDisease:AProspectiveCohortStudy.JAmCollCardiol.2016;66(14):1538-1548.doi:10.1016/j.jacc.2015.07.055.Saturated.
72. SacksFM,LichtensteinAH,WuJHY,etal.Dietaryfatsandcardiovasculardisease:ApresidentialadvisoryfromtheAmericanHeartAssociation.Circulation.2017;135.doi:10.1161/CIR.0000000000000510.
73. U.S.DepartmentofHealthandHumanServices,U.S.DepartmentofAgriculture.NutritionandYourHealth:DietaryGuidelinesforAmericans,1980.1980.https://health.gov/dietaryguidelines/1980.asp.
74. DietaryGuidelinesAdvisoryCommittee.ScientificReportofthe2015DietaryGuidelinesAdvisoryCommittee.;2015.
75. PhadkeVK,BednarczykRA,SalmonDA,OmerSB,HealthG.Associationbetweenvaccinerefusalandvaccine-preventablediseasesintheUnitedStates:Areviewofmeaslesandpertussis.JAmMedAssoc.2016;315(11):1149-1158.doi:10.1001/jama.2016.1353.Association.
76. SidneyS,QuesenberryCP,JaffeMG,etal.RecentTrendsinCardiovascularMortalityintheUnitedStatesandPublicHealthGoals.JAMACardiol.2016;1(5):594.doi:10.1001/jamacardio.2016.1326.
77. BarqueraS,Pedroza-TobiasA,Hernández-BarreraL,Bibbins-DomingoK,LozanoR,MoranA.Globaloverviewoftheepidemiologyofatheroscleroticcardiovasculardisease.ArchMedRes.2015;46(5):328-338.doi:10.1016/j.arcmed.2015.06.006.
78. PuskaP.Successfulpreventionofnon-communicablediseases:25yearexperienceswithNorthKareliaProjectinFinland.PublicHealMed.2002;4(1):5-7.doi:10.1136/bmjopen-2014-006070.
79. PietinenP,NissinenA,VartiainenE,etal.DietarychangesintheNorthKareliaProject(1972-1982).PrevMed(Baltim).1988;17(2):183-193.doi:10.1016/0091-7435(88)90062-X.
80. JousilahtiP,LaatikainenT,SalomaaV,PietilaA,VartiainenE,PuskaP.40-YearCHDmortalitytrendsandtheroleofriskfactorsinmortalitydecline:TheNorthKareliaProjectexperience.GlobHeart.2016;11(2):207-212.https://www.ncbi.nlm.nih.gov/pubmed/27242088.
81. VartiainenE.PersonalCommunication.2017.82. NicholsT.TheDeathofExpertise.Fed.January2014.
http://thefederalist.com/2014/01/17/the-death-of-expertise/.83. KatzD.OpinionStew.HuffingtPost.April2013.
http://www.huffingtonpost.com/david-katz-md/nutrition-advice_b_3061646.html.
63
84. NelsonL.WatchtherapidevolutionoftheAmericandietover40years,inoneGIF.May2016.https://www.vox.com/2016/5/18/11704458/american-diet-changes.
85. MossM.TheExtraordinaryScienceofAddictiveJunkFood.NYTimesMag.2017:1-25.http://www.nytimes.com/2013/02/24/magazine/the-extraordinary-science-of-junk-food.html.
86. KeysA,KeysM.EatWellandStayWell.GarcenCity:Doubleday&Company,Inc;1959.
87. KatzDL.SugarcoatingDietScience :SeekingSimpleTruth,PasttheFrosting.2017:1-3.https://www.linkedin.com/pulse/sugarcoating-diet-science-seeking-simple-truth-past-david?trk=mp-reader-card.
88. SilvermanM,FerenceB,ImK,etal.AssociationbetweenloweringLDL-Candcardiovascularriskreductionamongdifferenttherapeuticinterventions:Asystematicreviewandmeta-analysis.JAmMedAssoc.2016;312(12):1289-1297.doi:10.1001/jama.2016.13985.
89. Lloyd-JonesDM,MorrisPB,BallantyneCM,etal.2016ACCExpertConsensusDecisionPathwayontheRoleofNon-StatinTherapiesforLDL-CholesterolLoweringintheManagementofAtheroscleroticCardiovascularDiseaseRisk.JAmCollCardiol.2016;68(1):92-125.doi:10.1016/j.jacc.2016.03.519.
90. McGinnisJ,FoegeW.ActualcausesofdeathintheUnitedStates.JAmMedAssoc.1993;270(18):2207-2212.doi:10.1001/jama.1993.03510180077038.
91. KheraAV.,EmdinCA,DrakeI,etal.Geneticrisk,adherencetoahealthylifestyle,andcoronarydisease.NEnglJMed.2016;375(24):2349-2358.doi:10.1056/NEJMoa1605086.Genetic.
92. LeySH,HamdyO,MohanV,HuFB.PreventionandManagementofType2Diabetes:DietaryComponentsandNutritionalStrategies.Lancet.2014;383(9933):1999-2007.doi:10.1016/S0140-6736(14)60613-9.Prevention.
93. WangD,LiY,ChiuveS,etal.AssociationofSpecificDietaryFatswithTotalandCause-SpecificMortality.JAmMedAssoc.2016;176(8):1134-1145.https://www.ncbi.nlm.nih.gov/pubmed/27379574.
94. TrichopoulouA,BamiaC,TrichopoulosD.AnatomyofhealtheffectsofMediterraneandiet:GreekEPICprospectivecohortstudy.BrMedJ.2009;338(b2337).doi:10.1136/bmj.b2337.
95. SongM,FungT,HuF,etal.Associationofanimalandplantproteinintakewithall-causeandcause-specificmortality.JAmMedAssoc.2016;176(10):1453-1463.doi:10.1001/jamainternmed.2016.4182.
96. SpringmannM,GodfrayHCJ,RaynerM,ScarboroughP.Analysisandvaluationofthehealthandclimatechangecobenefitsofdietarychange.ProcNatlAcadSci.2016;113(15):1-6.doi:10.1073/pnas.1523119113.
97. FordE,BergmannM,KrögerJ,SchienkiewitzA,WeikertC,BoeingH.Healthylivingisthebestrevenge:findingsfromtheEuropeanProspectiveInvestigationIntoCancerandNutrition-Potsdamstudy.ArchInternMed.2009;169(15):1355-1362.doi:10.1001/archinternmed.2009.237.
98. EtemadiA,SinhaR,WardMH,etal.Mortalityfromdifferentcausesassociatedwithmeat,hemeiron,nitrates,andnitritesintheNIH-AARPDietandHealth
64
Study :populationbasedcohortstudy.BrMedJ.2017;357(j1957):1-11.doi:10.1136/bmj.j1957.
99. FreemanAM,MorrisPB,BarnardN,etal.Trendingcardiovascularnutritioncontroversies.JAmCollCardiol.2017;69(9):1172-1187.doi:10.1016/j.jacc.2016.10.086.
100. MichaR,PeñalvoJL,CudheaF,ImamuraF,RehmCD,MozaffarianD.Associationbetweendietaryfactorsandmortalityfromheartdisease,stroke,andtype2diabetesintheUnitedStates.JAmMedAssoc.2017;317(9):912-924.doi:10.1001/jama.2017.0947.
101. Sotos-PrietoM,BhupathirajuSN,MatteiJ,etal.Associationofchangesindietqualitywithtotalandcause-specificmortality.NEnglJMed.2017;377:143-153.doi:10.1056/NEJMoa1613502.
102. BernsteinAM,SunQ,HuFB,StampferM,MansonJE,WillettWC.MajorDietaryProteinSourcesandtheRiskofCoronaryHeartDiseaseinWomen.Circulation.2010;122(9):876-883.doi:10.1161/CIRCULATIONAHA.109.915165.Major.
103. DiabetesPreventionProgramResearchGroup.Reductionintheincidenceoftype2diabeteswithlifestyleinterventionormetformin.NEnglJMed.2002;346(6):393-403.doi:10.1056/NEJMoa012512.
104. LorgerilMDe,SalenP,MartinJ-L,MonjaudI,DelayeJ,MamelleN.MediterraneanDiet,TraditionalRiskFactors,andtheRateofCardiovascularComplicationsAfterFinalReportoftheLyonDietHeartStudy.Circulation.1999;99(6):779-785.doi:https://doi.org/10.1161/01.CIR.99.6.779.
FormoreinformationabouttheSevenCountriesStudy,visit:http://www.sevencountriesstudy.com/